A Text Sentiment Classification Method Enhanced by Bi-GRU and Attention Mechanism

Abstract

Text sentiment analysis is a natural language processing technique designed to identify the emotional tendencies expressed in text. In recent years, this field has garnered significant attention and is widely used in practical applications. For example, sentiment analysis is employed for brand reputation management on social media, public opinion monitoring, and risk control in fields such as finance, medicine, and politics. Sentiment analysis is also utilized in tasks such as personalized recommendation and natural language generation. Despite the numerous methods and techniques proposed and applied in text sentiment analysis research, challenges and problems persist. During the sentiment classification process, text data exhibits problems such as uncertainty and semantic diversity, noise, and errors, leading to low accuracy and efficiency of sentiment analysis models. To enhance sentiment analysis accuracy and efficiency, this paper proposes an improved text sentiment classification method based on Bi-GRU and self-attention mechanism. The attention mechanism is initially fused with the update gate of the Bi-GRU gating unit to obtain important feature information in the text content. Subsequently, the Bi-GRU is followed by a self-attention mechanism to perform secondary screening on the text features, and the softmax function is applied to text vectors for sentiment classification, significantly enhancing the accuracy of sentiment classification. The proposed method is tested on the public dataset Yelp Dataset Challenge, and the experimental results indicate a considerable improvement in the accuracy of sentiment classification.